Proposed Approach forOPAL Workflow

Sept 2009

Workflow Overview

Modules

Register orthogonal datasets &Apply transform

• Affine registration of sagittal and coronal data to axial data using mutual information (markers not currently used)

Resample datasets into new image

• Datasets “merged” by resampling• Weighting of each dataset based on distance

of point to nearest slice

Atlas model creation throughexpert segmentation

• Final model includes volumetric mesh and connected points representing muscle fibers

Segmentation by coarse user-controlled registration (initialization)

Segmentation by automatedimage-based registration

Specifications

General

• Quantify results using– Synthetic data: basic shapes, synthetic med data– Human data: ideal control data, patient data

• Prototyped using Matlab unless specified• Module input adaptable to allow for use of

Livewire

Assumption:Atlas model has corresponding MRI

• Alternatives if atlas has no corresponding MRI

Future Work

1. Collaboration with Yohann and Marek• Use of their workflow for Mesh-Match-and-

Repair for patient model segmentation• Contributions to this work:– The deformation of interior nodes, muscles and

landmarks (not just surface mesh)– Reliability constraints on landmarks, curves and

surfaces

Future Work

2. Possible new tongue structure• Surface mesh is arbitrary• Tongue consists of muscle fibers and skin is

for visual and collision calculation purposes only, appearing as a “skin” over muscles

• Registration deforms the muscle nodes only and ignores visible tongue surface